Location and Object Aware Model for Parallel Activity Recognition in Multi-Resident Smart Homes

Smart homes enable elderly individuals and people with impairments to live independently through remote monitoring of their activities. Sequences of sensor activations are mapped with their associated labels to recognize different activities. Activity recognition in a multi-resident environment is challenging due to multiple activities performed by different residents in parallel. A novel multi-resident activity recognition approach is proposed to separate the sensor events based on their location. A spatial matrix is generated to capture the spatial and temporal patterns of the activities, and activations of sensors are recorded as binary values. The spatial matrix is converted into images to represent parallel activities as multiple objects within each image. Activities are annotated within each image based on their location and sequence patterns. The YOLO (You Only Look Once) model is deployed to recognize parallel activities from the activity images. Experimental evaluation on the CASAS dataset shows a performance improvement of 7 percent compared with existing state-of-the-art approaches.